You are on page 1of 18

PHYSICAL REVIEW B 81, 205311 ͑2010͒

Simplified density-matrix model applied to three-well terahertz quantum cascade lasers
E. Dupont,* S. Fathololoumi, and H. C. Liu
Institute for Microstructural Sciences, National Research Council, Ottawa, Ontario, Canada K1A 0R6 ͑Received 17 December 2009; published 12 May 2010͒ A simplified density-matrix model describing the population and coherence terms of four states in a resonant phonon scattering based terahertz quantum cascade laser is presented. Despite its obvious limitations and with two phenomenological terms—called the pure dephasing time constants in tunneling and intersubband transition—the model agrees reasonably well with experimental data. We demonstrate the importance of a tunneling leakage channel from the upper lasing state to the excited state of the downstream phonon well. In addition, we identify an indirect coupling between nonadjacent injector and extractor states. The analytical expression of the gain spectrum demonstrates the strong broadening effect of the injection and extraction couplings. The gain is decomposed into three terms: a linear gain and two nonlinear components related to stimulated anti-Stokes scattering processes. The nonlinear gain is not negligible at high temperature. Under certain approximations, analytical forms of population and coherence terms are derived. This model is well suited for structures with only a few states involved. This model can simplify the optimization process for new laser designs; it is also convenient for experimentalists to adopt. DOI: 10.1103/PhysRevB.81.205311 I. INTRODUCTION PACS number͑s͒: 73.40.Gk, 73.23.Ϫb, 42.55.Ah, 72.20.Dp

The temperature performance of terahertz ͑THz͒ quantum cascade lasers ͑QCL͒ has been continuously improved to the point that the empirical limit of the maximum thermal energy, kBTmax, to the photon energy1 has been slightly surpassed in devices emitting from 1.2 to 3 THz,2,3 which makes these photonic devices unique. The maximum operating temperature reached so far is 186 K,4 and it is now only a step away from temperatures achievable by thermoelectric coolers. This will then open a broad range of applications from biological sensing, pharmaceutical sciences, THz wave imaging to hazardous materials detection, to cite just a few.5 These high-temperature devices are based on resonant longitudinal-optical ͑LO͒ phonon scattering for depopulation, except for the structures emitting close to 1 THz. These structures offer the advantage of consisting of only a few quantum wells ͑QWs͒, and therefore only a few relevant states per period are involved—which simplifies the model presented in this work. QCLs have attracted the attention of many theoreticians as well, and several useful models have been developed to predict the transport, the electron distribution, the gain, and the relative importance of different scattering mechanisms. We refer to the self-consistent rate equation model,6 the Monte Carlo simulations,7–10 the hybrid density-matrix Monte Carlo model,11 and the quantum theories based on nonequilibrium Green’s function12–16 or on density matrix.17–20 Some groups have reported a remarkable agreement between the theoretical electrical characteristics and experiments.14–16,20 However, the implementation of such models is difficult and also computationally very demanding. Alternatively, Scalari et al. have employed a simplified density-matrix model for a five-well QCL emitting at 3.7 THz. This work, after a fitting procedure of the optical and electrical characteristics, estimated the upper lasing state lifetime and relaxation time constant in the injector region.21 These parameters are useful for subsequent optimization. Kumar et al. developed a similar model which is used as a guideline tool for designing resonant LO-phonon scattering

based QCLs,22,23 as well as an analyzing tool of the laser performance.24 With the same model, they also simulated the gain profile for different range of laser intensities.22 Recently, our group has introduced the three-well THz QCL ͑Ref. 25͒ and carried out several optimization studies, such as the effect of the injection26 and extraction27 barrier thicknesses. This design consists of placing an active double well with a vertical intersubband transition in a sandwich between two identical LO-phonon resonant wells. The tunneling interactions between these three wells occur through the extraction and injection tunneling barriers. Here, the active double well and its two phonon wells for injection and extraction form the so-called three-well structure, even though there are four quantum wells. Within the main four states of this laser structure, five tunneling processes occur and they all participate in the diagonalization when calculating the adiabatic wave functions of this system. The simplified density-matrix model of Ref. 21 cannot be immediately applied to our three-well design because of the numerous tunneling and leakage processes taking place. In this work, we propose to extend the simplified density-matrix model to this case, which involves four states and five tunneling processes. In essence, our model is simply a system of rate equations for population and coherence terms, in which the electrons in each subband behave the same ͑i.e., like a single particle͒, regardless of their kinetic energy. We study two cases: first, a simple and very convenient case in which the gain profile is Lorentzian type and its strength is proportional to population inversion; and second, a more rigorous approach where the oscillating coherence terms between different states are directly related to the laser field. Near threshold, an analytical form of the gain can be derived; and it shows the strong broadening effect by the injection and extraction tunnelings. From this expression, three components can be identified: a main linear term proportional to population inversion and two nonlinear terms that are not negligible, particularly at high temperature. We apply this model to the experimental results of our injection barrier thickness study26 and find estimates for the electron temperature and tunneling pure dephasing time constant.
©2010 The American Physical Society


II. causes the lasing states to interact. the centered 50 Å of the phonon wells are Si doped at 7. Similarly. respectively. Upstream Phonon Well Active Double Well Downstream Phonon Well FIG. In Sec. V. more equations are given in the Appendix. The paper is organized as follows. a process called the wrong injection channel. These wave functions are then appropriately shifted in the growth direction and finally plotted as shown in Fig. the laser field just reduces the upper lasing state lifetime via stimulated emission. 1. we look at the three-well QCL design as an optically active double well interacting with an upstream phonon well for the carrier injection and a downstream phonon well for the extraction. and therefore. 25 at 12 kV/cm. 2–4.677. 26. tunnelings. However. we point out the principal limitations of our model. and 4 for the excited state of the downstream phonon well acting like an extractor level. which is three orders of magnitude smaller than the other four tunneling couplings. which denotes a high overlap between the wave functions. and extraction. 2. The relevant four states are named 1 for the injector level. the laser-induced coherence terms are added in the model. The laser field. From now. 25. The expected values of position for the lasing states. differ only by 1. The ⍀ij are the tunneling coupling strengths. 205311 ͑2010͒ 12 kV/cm Linj 48 20 73. III. 2. ͑Color online͒ Conduction-band diagram of the threewell QCL structure of Ref.4 nm. and ⍀24 for the wrong extraction channel. IV. Less importantly. In Sec. respectively. carriers injected to the upper lasing state can directly tunnel toward the extractor state. The reason is that only this pair of states is in close resonance with the laser frequency and has a significant dipole moment. In Sec. 25. The useful coupling strengths represented by green double-ended arrows are the injection.DUPONT. and 1–4. at low electric fields. but also between 1–3. AND LIU 161 PHYSICAL REVIEW B 81. this means they are isolated from the neighboring QWs. This approximation is treated in Sec. 25 at 12 kV/cm and the square modulus of the wave functions of the active double well and the upstream/ downstream phonon wells when taken isolated from the adjacent quantum wells. and an analytical expression for the gain close to threshold is derived and discussed.e. The subscripts ͑n + 1͒ and ͑n − 1͒ are referring to the downstream and upstream set of four states. i. 1. are represented by double-ended red arrows. The interactions considered in this model between the four states are schematically represented in Fig. ⍀14. These levels are part of the nth set of the states and the states belonging to the upstream and downstream sets are noted as i͑n−1͒ and i͑n+1͒. ⍀24. This is the reason why we say this design uses vertical intersubband transitions. The thickness in angstrom of each layer is given in vertically oriented font. Edsg. The various wave functions of these wells are computed separately.1 kV / cm for the structure of Ref. we describe the quantum system under study. close to ϳ12. 3͒. Active Double W. This figure represents the isolated wave functions of the three-well QCL reported in Ref.5 4(n-1) W14 Lext 42 161 96 96 4(n−1) 2 4 3 1 W12 D12 W13 Injection Barrier 48 1 t3 t2 WL 2 W24 D34 W34 3 Extraction Barrier 4 1 t4 t1 1(n+1) (n+1) Upstream Phonon W. Downstream Phonon W. EL. and we show that a wrong extraction channel is detrimental for continuous-wave ͑cw͒ operation. there is a weak direct parasitic coupling between the injector and extractor states. IV. From a simplistic point of view. Zii ͑i = 2 . 2 for the upper lasing state. At this electric field. In Sec. In Ref. the last section summarizes our conclusions. a process called wrong extraction. ⍀12. 3 for the lower lasing state. The last three couplings are not desirable for the laser. The main relaxation processes for achieving laser operation are drawn with solid wavy downward arrows and processes related to 205311-2 . V.2 ϫ 1016 cm−3 for an equivalent twodimensional carrier concentration N2D = 3. the oscillator strength between the lasing states is 0. ͑Color online͒ Schematic of the interactions considered in the model between the four relevant states in a three-well THz QCL.6 ϫ 1010 cm−2. The inelastic scattering of the carriers in various states are represented by wavy arrows. ⍀34. In Sec.. II. FATHOLOLOUMI. These channels are optimum at the design electric field of the laser. The laser field not only induces an oscillating coherence between 2 and 3. The injection ͑⍀12͒ and extraction ͑⍀34͒ are represented in green as opposed to the not so desirable tunneling channels like ⍀13 for the wrong injection channel. A parasitic and negligible channel ⍀14 between 1 and 4 can also occur. Finally. as will be discussed in Sec. FIG. ⍀13. as schematically represented by its Rabi frequency ⍀L = qZ23EL / ប in Fig. we solve the problem assuming the laser field only acts on the population of the lasing states. 2. the injector level can also interact with the lower lasing state. numerical applications of the model are performed to fit the experimental results of the injection barrier study of Ref. DESCRIPTION OF THE QUANTUM SYSTEM UNDER STUDY Initially. the model requires calculating the wave functions of different quantum wells that are coupled to each other by a slow tunneling process. As a result. VI. pure dephasing time constants as well as electron heating temperature are estimated. For readers who wish to use this model.

SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. VI may lead to an underestimation of ␶2. the wrong extraction coupling. This component contributes strongly to the linewidth of the optical intersubband transition. The vertical dashed line indicates the design electric field of the QCL. hot e−͒ is the LO-phonon scattering phonon energy. regardless of their kinetic energy. The same color and line style code for the different tunneling channels applies to both panels. Simulations have suggested a strong temperature effect on intersubband transition linewidth since the electron screening length ͑Debye length͒ of the impurity Coulomb potential increases with the electron temperature. ELO is the LO͑LO emi. ⍀24 = 2. is long. It is obvious that the electron heating temperature depends on input electrical power as well as the laser intensity. extractor. This might be a competing leakage path. Eij = Ei − E j. It is important to underline the fact that this model omits any thermal effects on the intrasubband component of the dephasing time constants between the states. nLO is the Bose-Einstein factor. ⍀34 is designed to be higher than ⍀12 to ensure population inversion even at high temperature. hot e−͒ ␶23 ͫ 1 E23 − ELO nLO + ͑LO abs͒ + ͑imp͒ . to an overestimation of the electronic temperature. 205311 ͑2010͒ phonon absorption are drawn by dashed wavy upward arrows. its value was kept independent of the electric field and lattice temperature. III. Since the THz QCLs are lightly doped. is lower due to the high confinement energy of the two states 1 and 3. ͑Color online͒ Panel ͑a͒ shows the detunings Ei − E j for the different tunneling processes between the four states. and a coupling strength. Panel ͑b͒ shows the coupling strengths ប⍀ij. The upper lasing lifetime.28 Each i → j tunneling is characterized by a detuning energy. 3.29 From this point on. tunnelings conserve the in-plane wave vector kʈ but are not subject to Pauli’s exclusion principle. the tunneling coupling strengths between the four states are derived by a simple tight-binding approach. ␶2. From these isolated wave functions.92 meV and the extraction coupling is ⍀34 = 1. T being the lattice temperature and ⌬Te the electron heating temperature. when the upper lasing time constant. we do not think this simplification will cause substantial errors. kB is the Boltzmann constant. in this work.39 meV. T + ⌬Te is the electron temperature on level 2.31–33 Moreover. On the other hand. ␶23 time for an electron on level 2 with a kinetic energy equal to ͑LO abs͒ is the LO-phonon absorption scattering ELO − E23. this component is called the pure dephasing time constant. The 1–4 coupling strength is only 0. With this model. our fitting exercise in Sec. in order to simplify our numerical applications.2– 0. and finally. At the design electric field.30 However. Microphotoluminescence experiments on resonant phonon scattering based QCLs have demonstrated that the temperature of lasing subbands can be ⌬Te Ϸ 100 K higher than the lattice. hot e−͒ ␶23 ͑2a͒ 205311-3 . ␶‫ء‬ ij. these sophisticated techniques should better predict the temperature dependence of the linewidth. ␶2. ␶2. particularly at low temperature. We also ignore Pauli’s exclusion principle for the intersubband lifetime of the states. versus the electric field. Our model consists of solving Bloch equations of four interacting states independently of the kinetic energy of the carriers. Since in our model. ͑LO emi. This means that the carriers on a particular subband behave the same.92 meV. The upper lasing state time constant. is simply modeled as ␶−1 2 = 1 + nLO exp ͑LO emi. ប⍀ij. Edsg. or equivalently. is stronger than the extraction coupling.3 ␮eV. ␶2. is thermally activated and. Therefore. and injector states are written in a similar manner ␶−1 3 = nLO . k B͑ T + ⌬ T e͒ ␶23 ͗␶23 ͘ ͑1͒ ͬ ͑imp͒ where ͗␶23 ͘ is the intersubband 2–3 unscreened impurity scattering time constant averaged over the MaxwellBoltzmann distribution of carriers on level 2. our solution is not resolved in the kʈ space of subbands. ␶2 carries most of the responsibility for the temperature degradation of the laser. We think that this approximation may be a serious one which simplifies the model to the great extent. which can be a severe limitation if the structures are heavily doped.56 meV. the injection coupling is ⍀12 = 0. ␶23 time for an electron at the bottom of subband 2 to level 3. The horizontal dashed line at zero detuning indicates the electric field for which the different tunnelings are in resonance. although we are aware that this simplification can distort the simulated current density versus electric field characteristic. This observation already suggests that this leakage path will not play a major role around the design electric field. Figure 3 plots the detunings and couplings for all five tunnelings. This model—as it is implemented here—concentrates the thermal effects only on the upper lasing time constant. like a single particle.34–36 However. The wrong injection coupling.35 The lifetime of lower lasing. ⍀13 = 0. LIMITATIONS OF THE MODEL FIG. is estimated from the 2–3 intersubband unscreened impurity and bulk LO-phonon scattering rates. simple models for the intrasubband scattering rate do not result in the same optical gain linewidth as the one computed in programs that use nonequilibrium Green’s function.

͑i = 2 . With these obvious limitations in mind. i. because the phonon absorption from the heavily occupied state 1͑n+1͒ to state 4 acts as a backfilling process at high temperatures. is given by the diagonal terms. ͑4͒. at high electric fields. ␻ 205311-4 . Interwell inelastic scattering between the different adjacent states was omitted because of the reduced overlap between these wave functions. ⑀0⑀rc 23 ͑5͒ where c / ng is the group velocity in the optical waveguide. The left matrix in the commutator of Eq. for the four-level system shown in Fig. wells per period. states belonging to different wells are only coupled to each other through coherent tunneling. For instance. Although these scattering terms should normally be included in the model. ͑2͔͒ to 4. ͑1͒ and ͑2͒. we believe this simplified model can give useful insights for the effects of the different couplings on the gain. In the next section. the time evolution of the density matrix. The electronphonon scattering rates are slightly overestimated since they are computed at resonance. except for 4͑n−1͒ → 2 where ␶4͑n−1͒2 Ϸ 7 ps. ␶sti sion of which is given by Eq. the most important term is ␶−1 1 .DUPONT. level 2 could be coupled to the second excited state of the next period active double well which is displayed by a light green wave function in Fig. the expresterms include the stimulated emission rate. For instance. Nevertheless. for a vanishing initial or final kinetic energy. We verified that the diagonalization of this matrix gives the same eigenfunctions as the adiabatic solution of the three-well structure. 205311 ͑2010͒ ␶−1 4 = 1 ͑imp͒ ͗␶41Ј ͘ + 1 + nLO ͑LO emi͒ ␶41 Ј + nLO ͑LO abs͒ . ͑3͒ where −1 ␶sti = c ␴op⌰ . we will model the steady-state populations and the static coherence terms without including the laser Rabi frequency which induces oscillating coherence terms between states. hot e ͒ appears in the expression of ␶−1 4 ͑␶2 ͒ Ј −1 23 −1 and ␶1 ͑␶3 ͒. ␶41Ј ͑2c͒ The terms with nLO as a coefficient are related to phonon absorption processes and are not negligible at high temperatures. It can also be extended to structures consisting of more. 3͒. MODEL WITHOUT LASER-INDUCED COHERENCE TERMS When the laser-induced coherence terms are excluded. and from 3 to 2.37 In the model. mitigated by dephasing processes. For the lasing states.e. Among the absorption terms in Eqs. their omission keeps the model simple. ␳22 − ␳33. are the reverse of the phonon emission processes 4 → 1͑n+1͒ and 2 → 3. IV. ͑3͒ represents the Hamiltonian of the two phonon wells and the active double well interacting with each other by tunneling. the interwell LO phonon scattering times are very long. Since this model only focuses on the four main states within one period. ␴op represents the optical intersubband cross section in square centimeter and reads as ␴op = ␲ q 2n g ͉Z ͉2␻L͑ប␻ − E23͒ . Z23 is the matrix dipole moment between the lasing states. ͑4͒. The decay of the populations. and yet accurate. ␳ii. 2. it cannot predict the leakage to more remote states. AND LIU PHYSICAL REVIEW B 81. ␶41Ј ͑2b͒ ␶−1 1 = nLO ͑LO emi͒ . ␳.. In Eq. Its simplicity makes the implementation very easy by experimentalists and can be used as a first-order optimization tool for new designs. This is the reason why the same term − LO emi −1 ␶41 ͑␶LO emi. these −1 . 1. We use Harrison’s numerical method to calculate them. can be written as d␳ 1 = dt ប − ΄΂ ΂ E1 ប⍀12 ប⍀13 ប⍀12 ប⍀13 ប⍀14 0 ប⍀24 E2 0 E3 ប⍀34 E4 ប⍀14 ប⍀24 ប⍀34 ΃΂ ␳11 ␳21 . The phonon absorption processes from levels 1͑n+1͒ ͓which is denoted 1Ј in Eq. ng ͑4͒ The variable ⌰ is the photon density in the cavity and ⌬␳ is the population inversion. FATHOLOLOUMI. The last matrix on the right-hand side represents the natural decay of population and coherence terms when these terms are placed in nonequilibrium condition. it is clear that the fitted parameters derived with our model should be treated with caution. We confirmed that the influence of this latter time is marginal. ␳31 ␳41 ␳12 ␳22 ␳32 ␳42 ␳13 ␳23 ␳33 ␳43 ␳14 ␳24 ␳34 ␳44 ΃΅ ␶−1 ʈ 14␳ 14 ␶−1 ʈ 24␳ 24 ␶−1 ʈ 34␳ 34 −1 ␶−1 4 ␳44 − ␶1 ␳11 −1 ␶−1 1 ␳11 − ␶4 ␳44 ␶−1 ʈ 12␳ 12 ␶−1 ʈ 13␳ 31 −1 ␶ʈ14␳41 ␶−1 ʈ 12␳ 12 −1 −1 ␶2 ␳22 − ␶−1 3 ␳33 + ␶sti ⌬␳ ␶−1 ʈ 23␳ 32 −1 ␶ʈ24␳42 ␶−1 ʈ 13␳ 13 ␶−1 ʈ 23␳ 23 −1 −1 ␳ − ␶ ␶−1 3 33 2 ␳22 − ␶sti ⌬␳ ␶−1 ʈ 34␳ 43 ΃ . or fewer.

etc. as in a tunneling process. . The first three terms represent the three tunneling currents departing from the injector state 1: injection ⍀12. ⑀kʈ. For the sake of simplicity.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. the static dipole is small. kʈ͉͘2 ϫ ␦͑⑀kʈ+qʈ − ⑀kʈ͒ . Therefore. 25.7 meV. is in the microelectron volt range. Tij. q is the elementary charge. the matrix element relative to a state in the phonon well. states 2 and 3 develop a static coherence by the tunneling trajectories 2 → 4 → 3 or 2 → 1 → 3. Lsp 23 ͑9͒ qʈ − ͗ j. despite the fact the parasitic coupling. and for a scattering potential. The differimpurity scattering example.5 kV/cm the adiabatic solution of the three-well structure shows a strong mixing between levels 1 and 4 with an anticrossing energy of 0.4. ͑6͔͒. ␳ij ͑i j͒. and hence the change in electric conductivity when lasing occurs. ␶ʈij. in the following discussion we consider the ionized impurity scattering from Si dopants introduced in the phonon well. where nm ij or ji. plus the “pure” phase randomization rate during −1 11 tunneling transport or optical intersubband transition. ͑9͒ contributes to ϳ5% of the total current. and ranges from −5 nm at 1 kV/cm to 2. This 205311-5 . In the design of Ref. The physical origin of last term which is directly related to the tunneling induced coherence between the lasing states is not as clear. the expression of current density. we see that the correlated intraband scattering rate between two states should substantially change whether the states are spatially separated. when levels 1 and 4 are both approximately equidistant in energy from the two lasing states. kʈ + qʈ͉V͑sc͉͒ j. . For instance. 205311 ͑2010͒ is the laser angular frequency. The solutions depend on different tunneling times between states i and j. We refer the reader to the Appendix for the explicit formulas of the approximate solutions of Eq. The model presented in this section does not calculate the laser-induced coherence terms. The fourth term is only a small correction which takes into account the different centroids of the two lasing states in the active double well. the emitted optical power. the last two terms of Eq. This is because several tunneling trajectories bridge these states. because the optical intersubband transition is vertical. The total phase loss rate between two interacting states i and j is half the sum of population decay rates via intersubband processes. which otherwise would not be coupled if leakage paths were absent. k ʈ + q ʈ͉V −1 band transition resonances with temperature will be minimal due to the thermal activation of ␶2. J. the remoteness of both wave functions from the ionized impurity potential and the strong overlap of the two wave functions imply that the two matrix elements in Eq. it assumes a voltage-independent Lorentzian function for the gain profile. We recall that the intrasubband component of the linewidth at a particular kinetic energy. The exponential decay of the coherence terms. kʈ͘. these terms are responsible for two non-negligible coherence terms between states 2 and 3 and between 1 and 4. terms in such a form ⍀ij␳nm. When writing the time evolution of the coherence terms in Eq. a different pure dephasing ‫ء‬ ␶ ‫ء‬. kʈ͘. we predict ␶23 ent total phase loss rates read 23 ‫ء‬ −1 ‫ء‬ ␶−1 + ͑2␶ j͒−1 + ͑1 − ␦23 ʈ ij = ͑ 2 ␶ i͒ ij ͒/␶ + ␦ij /␶23 . ប⍀14. This simplification is convenient to quickly estimate the total stimulated emission lifetime inside the cavity. These cross terms are responsible for the prethreshold shoulder in current-voltage characteristic that is observed in three-well designs ͑at ϳ8 – 9 V͒. ͑3͒. ͑3͒. −1 The higher ␶ʈij . such as ͗2 . ͑8͒ where ⌬ij = Eij / ប is the detuning in per picosecond. ͑6͒ and the ionized ‫ء‬ Ͼ ␶‫ء‬. which are defined as21 Tij = 2 1 + ⌬2 ij␶ ʈij 2⍀2 ij␶ ʈij .6 nm at 14 kV/cm. ͑7͒ where ␦ij stands for the Kronecker delta. instead. At most. ͑6͒ tend to cancel each other. is characterized by the total phase coherence time constant. such as ͗1 . and ⑀0⑀r is the permittivity of the active region. After some algebra. In the latter case.e. I͑ ¯ ͒ and R͑ ¯ ͒ are the imaginary and real components of a complex entity. ␶‫ء‬ ij . by the matrix element relative to a state in the active double well. kʈ͘ ij ͑⑀kʈ͒ = ␲ ͚ ͉͗i. V͑sc͒ = V͑imp͒. kʈ + qʈ͉V͑imp͉͒2 . The assumption is that the scattering potentials and matrix scattering elements for different pairs of states would be similar. wrong injection ⍀13.26. we assumed the same pure dephasing time constant for the five different ‫ء‬ tunneling channels ͑ij = 12. Lsp Ϸ Z11 − Z44 is the superperiod of the active region. to a lesser degree. can be written38 as ͑sc͒ ␶‫ء‬ ͉i. or overlapping each other. For instance. and the negligible parasitic ⍀14 channels.39 The tunneling times of the four main channels are comparable. on the order of 1–10 ps. at 8. we will see that this coherence is induced by the leakage paths and the so-called cross terms. . as in an optical intersubband transition in the active double well ͓see Eq. 13. is compensated. With this scattering potential. L͑ ¯ ͒ is a normalized Lorentzian function. . In the final steady-state solution. In the next paragraph. ␶−1 i. kʈ + qʈ͉V͑imp͉͒1 . with Eq. The first two terms are dominant in this expression. i. Analytical expressions of these quantities are given in the Appendix with some approximations. we notice the presence of cross terms between coupling strengths and coherence terms.͒: ␶‫ء‬ ij = ␶ . the more relaxed becomes the kʈ-conservation law. time constant is used between the two lasing states: ␶23 From the above discussion. In the former case. becomes J = 2⍀12I͑␳12͒ + 2⍀13I͑␳13͒ + 2⍀14I͑␳14͒ qN2D +2 −2 Z22 − Z33 ͓⍀24I͑␳24͒ − ⍀12I͑␳12͔͒ Lsp Z23 −1 ␶ʈ R͑␳32͒ .. which describes how long the phase correlation is conserved during the tunneling transport from state i to j. Z22 − Z33 is the static dipole between the lasing states. Having a temperature-independent pure dephasing time constant means that the broadening of tunneling or optical intersub- where N2D is the two-dimensional carrier density. j . ͑6͒ As an example. V͑sc͒.

The different leakage tunneling paths are artificially turned on or off to visualize their effects. 4. upper lasing state populations.4 without leakage to 8. The laser threshold electric field in this case increases significantly to 10. because the carriers injected to level 2 are quickly diverted to level 4 rather than been scattered to the lower lasing state: only few carriers accumulate on level 3. ␶2. The green lines with squares include the correct injection. and effect of the wrong injection tunneling.5 kV / cm͒ is close to experimental value. ⍀24. strong indirect coupling between the injector and extractor states will be a source of a leakage path. ⍀13. the population inversion is improved compared to the zero leakage situation. the indirect leakage at 8. The maximum population inversion is affected and hence the laser emission.85 ps͒ were inferred by the fitting exercise discussed later in Sec.40 Without leakage channels. the upper lasing state lifetime. Simulated current densities. is strongly reduced at higher temperature.5 kV/cm does not alter the population inversion significantly. and stimulated emission rates in the cavity are plotted in Fig. below 9. where leakage tunneling paths are artificially turned on or off. The magenta curve with crosses represents the experimental data at 4. whereas the effect of the wrong injection.DUPONT. and in optical intersubband transition is ␶‫ء‬ 23 = 0. The red lines with circles include the correct injection. 2–4. The rapid decrease in strength of this channel at higher electric field causes the laser threshold electric field to be increased slightly from 9. is turned on. Figure 4͑b͒ shows the depletion of the upper lasing state by the wrong extraction path. VI.3 kV/cm. 2–4.4 ps.e. 4͑c͒. ͑Color online͒ Simulation of ͑a͒ current density. Panel ͑a͒ shows that the current density around the design electric field increases significantly by the wrong extraction channel.2 K of a 0. The values used in the simulations for the electron heating temperature ͑80 K͒. The last panel shows that the wrong injection channel. below threshold. i.1 ϫ 1 mm2 metal-metal ridge waveguide laser with a structural design identical to that reported in Ref. 25 is plotted. 205311 ͑2010͒ FIG. 1–3. If only the wrong extraction channel. in order to visualize their impact.7 kV/cm and 1.85 ps. the transparency and the laser threshold electric fields are even more increased as a consequence of the depletion of the upper lasing state to the collector state. ͑1͒.2 kV/cm.4 kV/cm. 380 ° C rapid thermal annealing. 1–3. When the two wrong channels are included. as shown in Fig. by time-domain spectroscopy. It is unfortunate that we could not report an electrical measurement with Ohmic contacts from Ref. the set of blue lines with empty triangles includes the five tunneling paths. The maximum population inversion and laser emission are not affected by this channel. ͑c͒ population inversion ␳22 − ␳33 −1 ͑left vertical axis͒. the population inversion is simply given by ⌬␳͉⍀13=⍀24=0 = ␶2 − T34 − ␶4 T34 + T12 + 2␶2 + 2␶4 ͑10͒ and is represented by a green line in Fig.. 25’s wafer—as this material was exhausted—because the later “clone” wafer shows a higher threshold and a lower maximum operating temperature. This is consistent with the observation of strong THz absorption lines from the lower lasing state measured. mainly by the wrong injection channel. extraction and effect of the wrong extraction tunneling.1 to 9. 4. but loses its strength at the design electric field. Both contacts of this device were Ohmic by using an e-beam sputter deposition of Pd/Ge/Ti/Pt/Au with the thickness sequence of 550 / 1000 / 250 / 550 / 5000 Å. The lattice temperature is set at 10 K.4 kA / cm2 corresponds to lasing threshold. Nevertheless. The lattice temperature is 10 K. Before the transparency. extraction. is not as severe. the electrical characteristics at 4. AND LIU PHYSICAL REVIEW B 81. is set constant at 80 K. The consequence of this channel is to increase the transparency electric field from 7. FATHOLOLOUMI. the population inversion is strongly negative. However. For comparison. below transparency. However. It shows different cases. ⌬Te. the maximum population inversion is affected. 25. which results in a 14% decrease in the maximum population inversion at the design electric field. the presence of all tunneling channels is important to model the large current increase by the aforementioned 1–4 indirect coupling. the set of gray solid lines corresponds to the situation of no leakage paths. On the other 205311-6 . is very active at its resonance ͑5. below threshold.4 ps͒ and optical intersubband transition ͑0. so only ⍀12 and ⍀34 are nonzero.2 K of a metal-metal waveguide lasing device which structure is similar to the one reported in Ref. The kink ͑pointed by an arrow͒ of the current density at 10. our simulations with this simplified model are consistent with the experimental data: the order of magnitude for the peak current density is well predicted and the current density at the shoulder ͑ϳ8. the electron heating temperature. and stimulated emission rate ␶sti ͑right vertical axis͒ without the laser-induced coherence terms in the model. Finally. In the three panels. pure dephasing time constant in tunneling ͑0. 2–4. As mentioned in the discussion of Eq. followed by a 15 s. The blue line with solid triangles in panel ͑a͒ represents the simulated current density of a laser after threshold for a threshold population inversion of 10%. ͑b͒ upper laser state population ␳22. However at low fields. since the carriers falling on the extractor level are meant to be scattered efficiently to the next period injector level.6 kV/cm. population inversions. 4͑c͒.4 kV/cm͒. the pure dephasing time constant in tunneling is ␶‫ = ء‬0.

1 0. ͑7͔͒. Derivation of the gain When introducing the laser Rabi frequency. Therefore. level 4 is more populated than level 2. 205311 ͑2010͒ (a) 100 T (ps) 10 side. and should be minimized. 5 shows the subband populations in the nonlasing and lasing conditions. This observation brings an additional argument for a three-well QCL design based on diagonal transitions which demonstrated the record Tmax. In addition. V. the same panel shows that the four tunneling times are comparable within an order of magnitude. Panel ͑b͒ shows the populations of all the states and the population inversion on a nonlasing device ͑solid lines͒ and on a lasing device with a threshold population inversion of 10% ͑dashed lines͒. 4. From the above discussion. ⍀L. ͑A2͒–͑A6͒ in the Appendix.5 kV/cm. ␶2. where all four states are coupled to each other to form the 1–4 indirect resonance. ͑Color online͒ Panel ͑a͒ shows the four main tunneling times Tij as defined by Eq. ͑8͒. and that the wrong extraction channel increases the current density and slightly reduces the maximum population inversion. T13. This bottleneck at the depopulation stage explains the degradation of population inversion with temperature. we give approximate formulas that are derived from Eqs.23 ␶eff. T24 is about ϳ4 ps. and this explains the small finesse ͑Ϯ1.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. 5͑a͒ at ϳ5. in this particular design.2 ρ44 0 2 4 6 8 10 Electric field (kV/cm) 12 14 FIG.. In other words. Near the resonance of the wrong extraction channel ͑4. ␳11.4 ρ22 0.5 kV/cm. The transit time reads as ij T13 1 T24 0. it cannot compensate for the thermally activated ␶2.6 Populations ρ11 ρ33 ∆ρ 0. At the design electric field. while T13 is larger than 100 ps and hence cannot have a large influence. In the lasing condition.3 = T34 + ␶4. Figure 5͑a͒ shows the four main tunneling times versus electric field. GAIN PROFILE MODIFIED BY FOUR LASER INDUCED COHERENCE TERMS A.23 The objective of this section was to visualize the transfer of charges versus the electric field with and without the laser field. The range of electric fields for which these tunneling times are comparable is narrow. We also showed that the 1–4 indirect coupling is only a significant source of leakage at its resonance. in the Hamiltonian. ⌬␳th.5 kV / cm͒ of the simulated indirect resonance of ͉␳14͉. we concluded that the wrong injection channel. The tunneling resonance T24 will be broader than T13 and reasonably fast at the design electric field. we will see how the gain profile is strongly modified by the insertion of the laser interaction. To illustrate the effect of the wrong extraction path near the design electric field. Around 8. Of course. ͑5͔͒. at the design electric field. in the time evolution equation of the density matrix. Simulations are performed with the same parameters as Fig. does not work in this mode. and also. ␶−1 ʈ 24. hot e−͒ increases. We see that all four populations fluctuate around 8. quickly recovers from the resonance of the wrong injection.6 kV/cm͒. i.e. shown in Fig. because of the high coupling strength ⍀24. ␶transit = T24 ͑T34 + T12 + 2␶2 + 2␶4͒ T24 + ␶2 + ͑T12␶2 + T34␶4 + 4␶2␶4͒ T24 + ␶2 ͑11͒ and the product between the population inversion and the transit time becomes ⌬␳␶transit = T24 T34␶4 ͑␶2 − T34 − ␶4͒ − . which is limited by the phonon resonant scattering ␶4 ͓see Eq. In other words.4 kV / cm. this simple expression shows that the population inversion becomes less tem͑LO emi. is weakly temperature dependent by the nature of the resonant phonon depopulation scheme. ⍀L. does not severely limit the laser operation.6 kV/cm. the latter leakage channel is detrimental for cw operation and explains why the three-well design. the depopulation scheme cannot sustain the increasing flow of carriers originating from the shortening of the upper lasing lifetime with temperature. The model has used a voltage-independent Lorentzian gain profile ͓L in Eq. Therefore.41 The channel 2–4 is more to blame than channel 1–3. a rather low threshold current density4 and even cw operation. the threshold population inversion. 5. the “effective lifetime” of the lower lasing state. We see that the population of the injector state. T24 + ␶2 T24 + ␶2 ͑12͒ The product ⌬␳␶transit can be viewed as the current efficiency for population inversion. at low temperature. in which the factor T24 / ͑T24 + ␶2͒ represents the detrimental shunting effect by the wrong extraction channel. because of the high total phase loss rate of this tunneling. we obtain 205311-7 . which means that the depletion 2 → 4 will be still active. due to a combination of a strong resonance ͑high ⍀24͒ and other channels that also supply the extractor state. This can be perature sensitive as ␶23 achieved with diagonal 2–3 transitions. the upper lasing state population does not recover fully from the 2–4 resonance taking place at 4. with the assumption of an infinitely long tunneling time T13. Panel ͑b͒ of Fig. with vertical transitions. is fixed at 10% of the total number of carriers per period. which makes the building of a coherence between 1–4 possible. if possible.8 (b) T12 T34 0. the effect of this shunt channel is more pronounced at higher upper lasing state lifetime. In the next section.

͑17͒ from which the gain/absorption coefficient. which implies that our solutions are valid below threshold and slightly above. 205311 ͑2010͒ d␳ 1 = dt ប − ΄΂ ΂ E1 ប⍀12 ប⍀14 ប⍀12 E2 ប⍀24 ប⍀13 E3 ប⍀34 ប⍀14 ប⍀34 E4 ⍀L cos͑␻t͒ ប⍀24 ប⍀13 ⍀L cos͑␻t͒ ΃΂ ␳11 ␳21 .34 ʈ 23 ӷ ⌬ 23 + ␻ . Numerical applications of gain spectra Gop = ͩ ͪ ͑18͒ To illustrate the effects of injection and extraction tunnelings on the gain. Keeping in mind that in Sec. when we bring ⍀L to zero. As will be briefly discussed in the Apemission rate. for two cases. the right-hand ΂ ΃΂ ͑0͒ ␳12 0 ⌫13 − ⍀34 ⍀12 0 = ϫ ΂΃ ˜ ␳13 ˜ ␳14 ˜ ␳23 ˜ ␳24 ΃ ͑15͒ ͑16͒ . if ⌬12. ⑀ 0⑀ rc ប ⍀ L/ 2 B. we consider a voltage-independent Lorentzian 205311-8 . ͑3͒. ␶sti pendix. Provided the quality factor of the 2–3 resonance is high enough and the injection and extraction detunings are small. the population and coherence terms are calculated for a vanishing Rabi frequency. the rotating wave approximation is justified. The four oscillating coherence terms fulfill a simple system of linear equations ␳ij = ͓␻ Ϸ ⌬ij͔˜ ␳ije − ␻t + 0͒ ␳͑ ij . we find that the static terms of the density 0͒ matrix. − ⍀34 ⌫14 0 ⍀12 ⍀12 0 ⌫23 − ⍀34 0 ⍀12 − ⍀34 ⌫24 ͑13͒ The laser field which oscillates at the angular frequency ␻. where ⌫ij is the complex detuning of states i and j with respect to the laser frequency ⌫ij = ͑⌬ij − ␻͒ − The solution of Eq. FATHOLOLOUMI. Kumar22 showed numerical solutions of gain broadening by a strong stimulated emission rate in a three-level system with one injection tunneling channel. ˜ ␳ij is the laser-induced coherence between the same states and the bracket is a logical operator which equals 1 ͑0͒ if the expression inside is true ͑false͒. we computed the maximum gain versus the thicknesses of injection and extraction barriers. ␳31 ␳41 ␳12 ␳22 ␳32 ␳42 ␳13 ␳23 ␳33 ␳43 ␳14 ␳24 ␳34 ␳44 −1 ␶−1 1 ␳11 − ␶4 ␳44 ␶−1 ʈ 12␳ 12 −1 ␶−1 2 ␳22 − ␶3 ␳33 ␶−1 ʈ 13␳ 13 ␶−1 ʈ 23␳ 23 −1 ␶−1 3 ␳33 − ␶2 ␳22 ␶−1 ʈ 14␳ 14 ␶−1 ʈ 24␳ 24 ␶−1 ʈ 34␳ 34 −1 ␶−1 4 ␳44 − ␶1 ␳11 ␶−1 ʈ 12␳ 12 ␶−1 ʈ 13␳ 31 ␶−1 ʈ 14␳ 41 ␶−1 ʈ 23␳ 32 ␶−1 ʈ 24␳ 42 ␶−1 ʈ 34␳ 43 ΃΅ ΃ . ͑15͒ is found as ␶−1 ʈ ij . As expected. IV. ͑A1͒ ͑which models the static terms ␳͑ ij ͒ is changed and no simple analytical solution can be derived. 2 2 ˜ ⌫24 ⌫13 − ⍀34 ⌫13⌫14⌫24 − ⍀12 ␳23 ͑0͒ ͑0͒ = ͑␳22 − ␳33 ͒ 2 2 2 2 2 ⍀ L/ 2 ͑⍀34 − ⍀12͒ − ⍀34͑⌫13⌫14 + ⌫23⌫24͒ − ⍀12͑⌫13⌫23 + ⌫14⌫24͒ + ⌫13⌫14⌫23⌫24 ͑0͒ ⍀12 + ␳12 2 2 − ⍀34 − ⌫14⌫24 ⍀12 2 2 2 2 2 ͑⍀34 − ⍀12 ͒ − ⍀34 ͑⌫13⌫14 + ⌫23⌫24͒ − ⍀12 ͑⌫13⌫23 + ⌫14⌫24͒ + ⌫13⌫14⌫23⌫24 2 2 ⍀34 − ⍀12 − ⌫13⌫14 2 2 2 2 2 ͑⍀34 − ⍀12 ͒ − ⍀34 ͑⌫13⌫14 + ⌫23⌫24͒ − ⍀12 ͑⌫13⌫23 + ⌫14⌫24͒ + ⌫13⌫14⌫23⌫24 ͑0͒ + ␳34 ⍀34 . Here. ⍀L ͑0͒ ͑0͒ 2 ␳22 − ␳33 − ͑0͒ ␳34 ͑14͒ 0͒ where ␳͑ ij is the static tunneling induced coherence between states i and j. or in other 2 + ␶−1 words. fulfill the Eq. with a vanishing stimulated −1 = 0.DUPONT. ␳͑ ij . can be derived as ˜ N3Dq2͉Z23͉2ng␻ ␳23 I . we found static coherence terms between any combination of states in the absence of laser field. AND LIU PHYSICAL REVIEW B 81. we look for solutions in the form of 0͒ side of Eq. can potentially induce oscillating coherence terms between states that have energy spacing close to ប␻. if the laser intensity is not negligible. Gop. First.

where D is either A ͑antisymmetric͒ or S ͑symmetric͒. Figure 8 gives a simple picture of the mechanism explaining the voltage dependence of the peak frequency of the gain. and ␶23 maximum population inversion is obtained for ϳ33 Å thick barriers for the first case. The phase between these states is considered uncorrelated. ͑17͒. ‫ء‬ = 0. In panel ͑b͒ the complete gain model of Eq. The full model recommends limiting the tunneling interferences by increasing the barriers. Of course. The line with crosses represents the position of E23 as a function of electric field ͑quadratic Stark effect͒. 6. The results of this calculation are plotted in Fig.85 ps.4 ps. as well as the peak position and the two half width at half maximum points. even if these larger barriers decrease the population inversion ͓as in Eq. ͑Color online͒ Simulation results for the maximum gain ͑in per centimeter͒ as a function of injection and extraction barrier thicknesses with ␶‫ = ء‬0. 6 for ␶‫ = ء‬0. the four transitions have comparable strengths and the resonance energies are on both sides of E23: the total gain is centered around E23 and is broad. They can be written as 200 60 12 50 10 0 Lext (Å) 11 50 10 0 9 10 0 12 Photon Energy (meV) 14 16 18 20 FIG. and the oscillator strength mated like ␶−1 ʈ DDЈ Ј 23 would be proportional to the modulus square of the product of the components along states 2 and 3. they show a 5 meV full width at half maximum. is fixed at 80 K. 7. the peak gain is blueshifted before the design electric field. The unit of gain is per centimeter. ij −1 ͑21͒ This equation suggests that the tunneling pure dephasing −1 rate. is modeled as −1 S 2 A 2 ␶D = ␣ij␤ij␶‫ ء‬+ ͑␦D ␤ij + ␦D ␣ij͒͑2␶ j͒−1 . This model for the lifetime of the hybridized states is only qualitative. Aij = ␤ij͉i͘ − ␣ij͉ j͘ .5 kV/cm. contributes largely to the total broadening when the states are strongly mixed near resonance ͑␣ij Ϸ ␤ij͒. Dij. the electron heating temperature. For large positive injection and extraction de- 205311-9 .375 THz full width at half maximum. also called dressed states. which can induce interferences in the gain/absorption spectrum. This very simplified model considers four possible transitions between the two doublets. the gain shows a peak at a photon energy higher than E23. Sij = ␣ij͉i͘ + ␤ij͉ j͘ . The broadest spectra are obtained around the design electric field.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… (a) 60 100 PHYSICAL REVIEW B 81. the gain spectrum is assumed to be a −1 voltage-independent Lorentzian with a ͑␲␶‫ء‬ 23͒ = 0. ͑Color online͒ Contour plot of gain spectra for different electric fields. The electric field dependence of the lifetime of a dressed state.85 ps. and redshifted after. Around the design electric field ͓panel ͑b͔͒. ␶‫ ء‬. The lattice temperature is 10 K. For the sake of comparison.4 ps. The white dashed lines represent the position the two points at half width at half maximum. Below the design electric field. and 23 = 0. ͑19͒ ͑20͒ Defining ␪ij by tan͑␪ij͒ = 2⍀ij / ͉⌬ij͉. 205311 ͑2010͒ Ehwhm− Epk E 23 (b) 50 E hwhm + 150 50 14 Linj (Å) 100 50 50 13 10 0 40 Electric Field (kV/cm) 150 100 50 50 30 30 40 50 60 30 40 50 0 FIG. are displayed in Fig. In panel ͑a͒. narrow linewidth symmetric Sij and antisymmetric Aij dressed states are formed by the mixing of states 1–2 on the injection side and states 3–4 on the extraction side. the pure dephasing time constant in tunneling is ␶‫ = ء‬0. ͑10͔͒. Relative to E23. the complete model ͓panel ͑b͔͒ suggests further isolating the active double well from the phonon wells. For long phase coherence time constants. S12 and DЈ = A34 . The white solid line represents the position of the peak gain. gain profile with a ប␶−1 ʈ 23 half width at half maximum. Second. ⌬Te. We are going to use a simple tight-binding model to illustrate the evolution of the photon energy at peak gain versus electric field.85 ps. ͑17͒ is used. T = 50 K. We also see that the peak gain with the complete model is significantly smaller than the one predicted with the Lorentzian model. The broadening of the optical intersubband transition between states D = A12 . a value between 2ប⍀34 and 2ប͑⍀12 + ⍀34͒. this means that the coherence ˜ ␳23 depends only on population ͑0͒ ͑0͒ ␳23 / ⍀L = ͑␳22 − ␳33 ͒ / ⌫23. and in optical intersubband transition is ␶‫ء‬ 23 = 0. the component ␣ij will be sin͑␪ij / 2͒ ͓or cos͑␪ij / 2͔͒ and the component ␤ij will be cos͑␪ij / 2͒ ͓or sin͑␪ij / 2͔͒ for positive ͑or negative͒ detuning. however. 7. Getting the correct extraction barrier is more critical than having the correct injection barrier. and above the design electric it is redshifted relatively to E23. as a thick extraction barrier can damage the depopulation mechanism. The full width at half maximum is 5 meV at 12. ⌬Te = 80 K. the same color scale is used in both panels. 2˜ complete model summarized in Eq. formed by the coupling of 1–2 and 3–4 states.4 ps. we use the inversion. ͉͗D ͉ 2͗͘3 ͉ DЈ͉͘2. S34 would be esti−1 −1 −1 = ␶D + ␶D + ␶‫ ء‬. the density-matrix formalism takes into account the phase relation between the states. Panel ͑a͒ shows that the ⌬Te = 80 K. ␶‫ء‬ T = 50 K. The full gain spectra at 10 K for different electric fields. with ϳ45 Å barriers. At this bias.

are neglected. the nonlinear gain term is at maximum ence. we display the contribution of each term to the total gain of panel ͑a͒. On the upper and lower parts of the figure.14.43 in which the coherence between the two lowest levels. the first term on the right-hand side of Eq. 10. one finds the gain of a threestate system with one injection tunneling as demonstrated in Ref. The second term in the numera͑0͒ ͑0͒ − ␳22 ͒ is associated with a tor in Eq. where ⌫12 is the complex detuning at the injection side. the symmetric and antisymmetric states are represented by horizontal solid lines and the estimated linewidth of these states are represented by vertical oriented Lorentzian peaks. AND LIU (a) 11 kV/cm (b) 12 kV/cm A S A 12 PHYSICAL REVIEW B 81.4 ps. 205311 ͑2010͒ (c) 13 kV/cm plings. and neglecting the same nondesirable tunneling couplings. and in optical intersubband transition is ␶‫ء‬ 23 = 0.14. and if the nondesirable tunneling cou- where ⌫34 is the complex detuning at the extraction side. is caused by an external pump laser field ͓see Fig. 9͑c͔͒. It is interesting to note that this latter expression looks similar to the gain of a three-level system in the p configuration. which are not linked to the population inversion. the contribution of each transition is represented by a dashed colored Lorentzian line and the sum of four transitions by a solid black line. This system has been extensively studied for its potential in amplification without population inversion. It is the dominant component and represents a strongly modified Lorentzian gain profile under the effect of coherent injection and extraction. In our lasers. is prepared by a pump laser ͓see Fig. ͑17͒. ͑17͒. the electron heating temperature. ␳33 around ប␻ Ϸ E24.4 ps. ͑22͒. simply modeling the “sum” of p and h configurations. on the gain ͑0͒ ͑0͒ ͑0͒ ͑0͒ − ␳33 . ⍀13. FATHOLOLOUMI.43 Recently. ⍀13. The lattice temperature is 10 K. ͑17͒ represents the stimulated scattering from the upper lasing state 2 to the extractor level 4 in presence of the “extraction field” ⍀34. Also.85 ps. The effects of the nondesirable tunneling couplings. blue for S12 → A34. green for A12 → S34. In panels ͑b͒–͑d͒ of Fig. ͑23͒ C. the only difference comes from the injection tunneling Rabi frequency which drives the coherence between states 1 and 2. ␳11 maximum around ប␻ Ϸ E13 and is highly dispersive. 8. For positive popu͑0͒ ͑0͒ − ␳22 Ͼ 0. ͑b͒ around. On the upper part of the figure. On the lower part. A large positive ͑negative͒ extraction detuning decouples the final state A34 ͑S34͒ from the short-lifetime state 4.DUPONT.44 The gain of this Raman laser is modeled as in Eq. ⌬12 − ␶−1 ʈ 12. the second term of Eq. are called the nonlinear components of the gain. and this explains why the spectrum is narrower below or above the design electric field. ͑22͒ is derived as ˜ ␳23 = ⍀ L/ 2 ͑0͒ ͑0͒ ͑0͒ ͑0͒ − ␳33 ͒ − ͑␳33 − ␳44 ͒ ͑␳22 2 ⍀34 ⌫34⌫24 Gain (arb. ⍀2 ⌫23 − 34 ⌫24 . ␳12 .24. an expression very similar to Eq. and ͑c͒ after the design electric field. is fixed at 80 K. The expression of Eq. and brown for S12 → S34. ␶23 205311-10 . the pure dephasing time constant in tunneling is ␶‫ = ء‬0. The two other terms. ͑17͒ can be identified as a third-order process. lasing by an intersubband Stokes Raman process was demonstrated inside a midinfrared ͑mid-IR͒ QCL acting like a pump. ͑17͒ is. the oscillator strength is concentrated mainly on S12 → A34 transition which is slightly redshifted with respect to E23. and ⌬Te = 80 K. ͑22͒ 0 10 20 Photon Energy (meV) 10 20 10 20 FIG. 9͑b͔͒. 1 and 2. If are T = 10 K. 42. ␶‫ = ء‬0. For large negative detunings ͓panel ͑a͔͒. which resonantly scatters carriers from the injector level 1 to lower lasing state 3 in presence of the “injection field” ⍀12. Different components of the gain Equation ͑17͒ can be decomposed into several terms that are associated with different physical processes. units) tunings ͓panel ͑c͔͒. ⌬34 − ␶−1 ʈ 34. in which the coherence between the two highest levels. ͑17͒. the transition A12 → S34 is dominant and is blueshifted. and ␳34 . 3 and 4. ⌬Te. If the extraction coupling ⍀34 and the extractor state 4 are ignored in Eq. this nonlinear gain term is at lation difference. where two coherence terms between two highest and two lowest energy states are electrically driven by tunneling ͓see Fig. ͑22͒ ͑proportional to ␳11 stimulated anti-Stokes scattering process. when ignoring ⍀12 and the injector level 1 in Eq. are hidden in ␳22 We find it instructive to show the relative strength of the three terms in Eq. ˜ ␳23 = ⍀ L/ 2 ͑0͒ ͑0͒ ͑0͒ ͑0͒ − ␳33 ͒ − ͑␳11 − ␳22 ͒ ͑␳22 2 ⍀12 ⌫12⌫13 12 34 S34 ⍀2 ⌫23 − 12 ⌫13 . This expression reminds us the gain of a threelevel system in the h configuration. Similar to Ref. ͑17͒ ͑proportional to population inversion͒ is called the linear component of the gain. The third term of Eq. For positive population differ͑0͒ ͑0͒ − ␳44 Ͼ 0. the same color code scheme is used: red for A12 → A34. The vertical dashed line shows the position of E23.85 ps. From this point on. ͑Color online͒ Simplistic tight-binding model of the gain spectrum ͑a͒ before. The input parameters used during the simulation ‫ء‬ = 0. 42. The uncoupled states 2 and 3 are represented by horizontal dashed lines. therefore.24. 9͑a͔͒.

e. ͑Color online͒ Contour plots of the total gain and its three components. where the coherence between the two highest states is conditioned by a field ͑laser. ͑Color online͒ Four contour plots of the total gain ͑a͒ at 10 K.. the peak gain is blueshifted with the electric field ͓Fig. Our numerical results confirm that the nonlinear terms of the gain are more active in THz than in mid-IR QCLs because of the relatively long dephasing time. 11. measurements of the voltage dependence of spectra for different temperatures might be able to confirm the contribution of the nonlinear gain. The maximum value of the sum of two nonlinear terms is about ϳ10% of the maximum linear gain. the third term is about one third of the second term because of the smaller coher͑0͒ ͑0͒ Ͻ ␳12 . At 10 K. the total gain is characterized by a negative Stark effect. ͑c͒ Schematic representation of a four-level system. the peaks of the linear and nonlinear gains coincide in energy. The thin white line in panel ͑a͒ shows the position of the peak gain versus electric field.9 ͑13. 9. where the coherence between the two lowest states is conditioned by a field ⍀34. 12 10 9 10 15 20 10 15 20 −50 Photon Energy (meV) we compare the total gain ͓panel ͑a͔͒ and the linear gain ͓panel ͑b͔͒. 11͑a͔͒ because of the stronger contribution of nonlinear terms. as in Fig. while the population inversion is negative. the gain is positive. we see that the nonlinear gain of the second term ͓panel ͑c͔͒ has an impact. the coupling strengths ⍀12. 205311-11 . the relative contribution of the two nonlinear gain terms increases significantly. Unlike the mid-IR QCLs. 205311 ͑2010͒ (a) (b) 2 1 2 12 30 20 10 3 3-level system “p-configuration” (a) 3 W34 4 3 4 Electric Field (kV/cm) 10 9 14 (d) (c) 0 −10 −20 −30 −40 3-level system “h-configuration” (b) 4-level THz QCL (c) FIG. This observation suggests that the nonlinear components can potentially change the total gain spectrum ͑shoulder or shift of the peak frequency͒ when levels are not aligned. 11 we show that. 10. Electric Field (kV/cm) 100 10 9 14 (d) (c) 80 60 40 20 0 10 9 10 −20 15 20 10 15 20 12 Photon Energy (meV) FIG.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… 2 1 W12 14 PHYSICAL REVIEW B 81. Generally. 0͒ ͑0͒ The first term depending on ͑␳͑ 22 − ␳33 ͒ is displayed in panel ͑b͒. 24͒ and therefore. At the design electric field.42 In THz QCLs. i. showing all three components as decomposed in Eq.34 are comparable to the dephasing rates ␶−1 ʈ ij ͑ ij = 12.5͒ kV/cm and for photon energies lower ͑higher͒ than E23. but the linear gain has a slight negative Stark effect ͑as in Fig. but at T = 140 K. but the relative contribution of nonlinear gain is actually higher below or above the design electric field. 34. Close to the maximum operating temperature. At 10 K. this enhances the contribution of the nonlinear gain. tunneling͒ with a coupling strength ⍀12. i. and the third term 0͒ depending on ␳͑ 34 in panel ͑d͒. when the population in14 (a) (b) 120 12 FIG. In Fig. while the nonlinear gain shows a strong positive Stark effect. below or above the design electric field. The dispersive nonlinear gain in panel ͑c͒ is strong enough to change the linear gain contour ͓panel ͑b͔͒ into a different total gain contour ͓panel ͑a͔͒. the THz QCL works in a regime where the nonlinear gain plays an important role. and the peak is mostly redshifted ͓Fig. a decrease in peak frequency with electric field. This term gives an additional anti-Stokes gain at the photon energy around E13 and is strongly dispersive. The unit of gain is per centimeter. version is not maximum. the ͑0͒ second term depending on ␳12 in panel ͑c͒.. an increase in peak frequency with electric field. i.. 7 are used in the simulations. ͑b͒ Schematic of h configuration. such as the three-well THz QCL which can be viewed as the “sum” of p and h configurations. 13. which comes from ⍀34 Ͼ ⍀12 and ␶4 Ӷ ␶2 and ence ␳34 ͑as shown in the Appendix͒. The constant gain lines at 0 cm−1 are displayed by a solid black line. Therefore. ͑17͒. as the temperature is raised to 140 K and the population inversion is lowered. the opposite situation is expected. The white line on panel ͑a͒ shows the positive Stark effect of the peak gain. the dephasing time is not limited by interface roughness scattering in THz devices. ͑Color online͒ ͑a͒ Schematic representation of a threelevel system in p configuration. The same parameters as in Fig. 10͑a͔͒ by the large contribution of the linear component. i. 10.e.e. This is small. 8͒. The same figure shows that below ͑above͒ 10.. At 140 K.e.

205311 ͑2010͒ −10 0 10 ¯ hω − E23 (meV) 0 0 1 −10 0 10 ¯ hω − E23 (meV) −10 0 10 ¯ hω − E23 (meV) 3 4 ¯ hΩ34 (meV) 2 FIG. this can be explained by the large broadening of the transitions and the small photon energy difference.23 Our analytical model can predict the conditions for a multipeaked gain spectrum. 2. and hence the transitions between each of these states to the third level can be distinguished. This calculation was performed for perfect injection and extraction resonance conditions. ͑22͒ and ͑23͒ will show a double-peak behavior when ˆ = ⍀12 Ͼ ⍀ 12 hΩ12 (meV) ¯ ͱ ͱ ␶−1 ʈ 13 ␶ʈ13 2+ ␶ʈ23 ␶−1 ʈ 24 ͑24͒ and ˆ = ⍀34 Ͼ ⍀ 34 ␶ʈ24 2+ ␶ʈ23 . These insets illustrate the broadening of linear gain and the decrease in the peak gain by the extraction coupling strength. due to the resonant ˆ . The plots displayed as insets of Fig. FATHOLOLOUMI. 12 shows that a double-peaked gain occurs before this simplistic criterion is met. ˆ . In the case where the resonance conditions are not perfectly met. 2. 23 explained. it comes from the anticrossing between 3 and 4. 8͑b͔͒.5 meV.3 = T34 + ␶4.8. This criterion is not accurate. the double-peak behavior to appear for larger ⍀ ij simulation results for nonaligned states show that the phase ˆ . Criterion for double-peaked gain means that the coherent injection or extraction is sufficiently strong to reveal the existence of its associated dressed states in the gain spectrum. COMPARISON OF THE MODEL WITH EXPERIMENTAL DATA In order to estimate the pure dephasing time constant and the electron heating temperature. as the coupling strengths become unbalanced as to favor a side. are also shown ͓Eqs.. The calculation is performed for perfect alignment of the states at the injection and extraction sides. However. Figure 12 shows the number of peaks of the imaginary part of the first term of Eq. The first two have a single-peaked behavior. ⍀ ij higher the coupling on the injection or extraction side. ͑Color online͒ The “phase” diagram of number of peaks in the spectrum of the imaginary part of the first term of Eq. ͑25͒ As Ref. ⍀ij. our three-well QCLs are designed for ⍀12 ϳ 1 meV and ⍀34 ϳ 2 meV. the coupling limit on the extraction side. 2͉⍀12 − ⍀34͉. In either of these cases. the ˆ on the other side is required to obtain higher the ratio ⍀ij / ⍀ ij a double-peaked linear gain. in a four-level system like the one represented in Fig. AND LIU 2 2 1 −10 0 10 ¯ hω − E23 (meV) 1 Ω12 Ω34 2 PHYSICAL REVIEW B 81. and the last one shows a double-peaked gain.e.23. coupling is absent. ␶eff. and from the fast “effective lifetime” for the lower lasing state. In the same reference. is larger than a limit value. between the two dressed state transitions A12 → A34 and S12 → S34 ͓Fig. ͑17͒ versus the extraction and injection couplings. and gain. 2. in resonance condition. More precisely. Typically. The limit values of the coupling strengths to obtain a double-peaked gain when the other ˆ . 12 show the spectra of the imaginary part of the linear ␳23 at ⍀12 = 1 meV. ␶‫ء‬ 23 = 0. Qualitatively. we further investigate the linear gain of a three-level system in the p and h configurations. therefore they should show a single-peaked linear gain. is larger than that of injection. within the framework of a simplified density matrix model. the gain spectrum shows a single peak. only the tunneling coupling at one of the barriers is considered. The number of peaks expected are indicated in square boxes. As expected. a convenient criterion was given for a double-peaked gain around the design electric field. This These equations impose a strong coupling between the interacting states to overcome the total phase loss rate ␶−1 ʈ 13 and ␶−1 ʈ 24.5. At ⍀34 = 2. and 3. From simplistic arguments.85 ps. diagrams become more complex than just an increase in ⍀ ij To understand this behavior. ͉⍀12 − ⍀34͉ reaches ⍀ 12 34 as Fig.5. The coupling strengths. The strong coupling ensures a bigger splitting between dressed states Aij and Sij. When two tunneling couplings from injector and extractor sides have comparable strengths. 8͑b͒ or 9͑c͒.8 meV. the large gain linewidth mainly originates from the extraction tunneling mechanism. the linear gain is at the boundary of having one or two peaks. coherence terms.8. we intuitively expect the ˆ . we compare the results of 205311-12 . ⌬12 = ⌬34 = 0.5 meV. ͑17͒ ͑which represents the linear gain͒ as a function of the injection and extraction coupling strengths. the double-peaked gain should then appear when ˆ +⍀ ˆ . ⌬12 = ⌬34 = 0. the third is marginally double peaked. ⍀ ij D. and for four different extraction term of ˜ couplings ⍀34 = 1. the sketches in the insets are plotted with the same vertical scale. One can demonstrate that the imaginary part of the linear gain in Eqs. and in the next section we will compare our model with published experimental data on three-well THz QCLs.DUPONT. We have here estimated populations.4 ps. Several groups have observed dual-wavelength operation of THz QCLs with modes 2. ͑24͒ and ͑25͔͒. and ␶2 = 2 ps. VI.5–3 meV apart. ⍀ ⍀ 34 12 phonon scattering from level 4.45 This has been attributed to the anticrossing on the injection side45 or on extraction side. ⌬12 0 and/or ⌬34 0. The parameters used are ␶‫ = ء‬0. and broadening of the gain by the extraction coupling is apparent. the double-peak behavior associated with either the injection or the extraction appears. which also depends on the extraction coupling. and 3. For comparison. This “phase diagram” shows that a double-peaked gain behavior occurs if at least one of the ˆ . i. Four examples of these spectra are given at ⍀12 = 1 meV for different extraction couplings ⍀34 = 1. 2. However. 12.

the 200-␮m-wide and 1-mm-long ridge lasers reported in the injection barrier study have been remeasured with shorter current pulses ͑200 ns͒ and with better control of the electrical reflections. For various pure dephasing time constants for the intersubband ‫ء‬ . might be underestimated. ͑panel a͒ and electron heating temperature. suggests that the model of the permittivity of the metals used in the waveguide calculation overestimates the loss. ␶‫ ء‬and ⌬Te are adjusted to minimize the standard deviation of the maximum gain with respect to the estimated total waveguide loss. our model assumes a homogeneous broadening of the optical and tunneling transitions. if the phonon absorption is arbitrarily removed from the model. On the other hand.6 THz gain linewidth on the same design as the one under study in this paper. which was estimated as ␣M + ␣W ϳ 40 cm−1. 4 confirms. The fitting procedure needs the total waveguide loss coefficient. the experimental maximum operating temperature points of Ref. the simulation applied to the diagonal structure in Ref.35 our fitting exercise suggests an intersubband pure dephasing time constant on ‫ء‬ ϳ 1 ps. 23. 205311 ͑2010͒ our density-matrix model with the experimental data on laser performance versus injection barrier thickness. the thick 400 nm top n+-GaAs contact layer doped to 3 ϫ 1018 cm−3. 7͒. Since the electron temperature of the lasing subbands can be as much as 100 K higher than the lattice.85 ps.4 ps. stants are forced to adapt themselves to this increase in gain linewidth and hence. the double-peak spectra at specified voltages. because a decrease in this parameter has to be choice of ␶23 compensated by a longer upper lasing state lifetime. the fitted pure dephasing time con- FIG. the minimum standard deviation with the fitted parameters ␶‫ ء‬and ⌬Te is about 9 – 10 % of the estimated waveguide loss. ␶‫ء‬ 23. the Ti/Pt layers were modeled in the waveguide calculation as one metal with a 70 000 cm−1 plasma frequency and a short 12 fs electron lifetime. The vertical errors bars indicate the deviation of ␶ and ⌬Te which induce a 10% degradation of the fit.4 ps and depends weakly on the choice of ␶23 The electron heating temperature is more affected by the ‫ء‬ .5 cm−1 constant gain line. and therefore to an underestimation of electron heating temperature. the maximum gain was computed for the different injection barrier thicknesses and maximum operating temperatures of the six series of devices studied in Ref. the simulated waveguide loss with a 4. the maximum gain using the fitted parameters derived form the injection barrier study. the simulated total waveguide loss of a thin 100 period structure. The results are given for 40 and 35 cm−1 total waveguide loss. Indeed. like in Ref. This high loss is due to the high doping concentration in the active region. 205311-13 . due to its high coupling strengths and small current. Provided all parameters but the number of periods are the same between the two wafers used in the injection and extraction studies.33 ps. the order of ␶23 We recall that from the beginning. we used 25 nm Ti/55 nm Pt/300 ͑1000͒ nm Au for the top ͑bottom͒ mirrors. This disagreement could be due to different quality of growth in each of the samples. the nonlinear gain is weaker than in the vertical design ͑as in Fig. ␶‫ء‬. For a given ‫ء‬ ␶23 . Figure 14 illustrates the fit with the parameters ‫ء‬ ␶‫ = ء‬0. If the simulation is performed as a function of the extraction barrier thickness with the same input parameters. As a result. As a result. Figure 13 shows the summary of the fitting procedure. we observed an increase in the maximum operating temperature on the samples with a thin injection barrier. ⌬Te. At 40 cm−1. ␶23 and electron heating temperatures. We fitted the experimental maximum operating temperature with our density-matrix model and compared the predicted threshold current density with the experiment. In addition. is cen‫ء‬ . the gain linewidth is ϳ1.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. ͑panel b͒ as functions of the pure dephasing time constant in optical ‫ء‬ transition. 26. and the nonoptimized metallic mirror of the double surface plasmon waveguide. 13.4 ␮m thin active region. the discrepancy between 64 cm−1. Williams46 has measured the electroluminescence linewidth on resonant-phonon depopulation devices and found it comparable to our predicted value. Jukam et al. ⌬Te. ͑Color online͒ Fitted tunneling pure dephasing time constant. This should be compared with 64 cm−1. the fitted electron heating temperature would be ϳ10 K higher and the pure dephasing time constant would be ϳ0. Filled ͑open͒ symbols are obtained with a 40 ͑35͒ cm−1 waveguide loss. Here. Compared to our data published in Ref.26 We could have chosen a comparison with the data versus extraction barrier thickness27 but this latter study is not as comprehensive as for the injection barrier. transition. The effect of backfilling by phonon absorption from 1͑n+1͒ to 4 is not negligible. the thin active region of the wafer used in the extraction barrier study could enhance the sensitivity of the fit to the THz permittivity of the lossy metals used in the surface plasmon waveguides. 27 align on the 55 Ϯ 2. tunneling pure dephasing time constants. ␶‫ء‬. ␶‫ء‬. With the fitted values of the pure dephasing time constants. and ␣ M + ␣W ϳ 40 cm−1. and 55 cm−1.40 measured a 0. 26. ␶23 = 0. The fit is not perfect but acceptable. the losses might be overestimated. In Ref.2 THz at the design electric field ͑as in Fig. the fitting procedure might lead to an overestimation of upper lasing state lifetime. Titanium being a poor conductor. 26. The tunneling pure dephasing time constant. In the fitting procedure. if there is a period-to-period inhomogeneity in our structure. With our fitted parameters. tered at ϳ0. 1͒. In this diagonal design.

7 cm−1. ͑Color online͒ Contour plot of the maximum gain ͑in cm−1͒ versus the thickness of injection barrier and lattice tem‫ء‬ perature for ␶‫ء‬ 23 = 0. VII. we compare the simulated threshold current density at 10 K. ACKNOWLEDGMENTS 2 500 10 K 0 30 40 50 inj 60 L (Å) The authors would like to thank M. 15. This problem could be minimized by removing one tunneling process.5 60 2 2. Equation ͑17͒ shows that the gain spectrum is significantly broadened by both tunnelings.5 01 between experiment and simulation is more pronounced for thick barriers. As the disagreement 2500 Threshold Current (A/cm ) 2000 Tmax 1500 1000 We have developed a simplified density-matrix model for a four-state resonant phonon scattering based THz QCL. Khurgin for fruitful discussions. the model showed its limits because of the large number of fitted parameters and its numerous simplifications. and at Tmax for 40 cm−1 total waveguide loss. (3) FIG.5 4. It describes the populations of different subbands and the coherence terms between them. this leakage path should be minimized. At 10 K. with the experimental values. independently of the current. If possible. The equivalent upper lasing state lifetime when the maximum gain is reached is given on the left-most vertical axis. The white dots plot the redetermined maximum operating temperature of the six series of devices reported in Ref. injection and extraction.45 In the expression of the optical gain.85 ps. i. FATHOLOLOUMI.4 ps. by using a resonant phonon scattering mechanism for injection47 or a two-well design.6 THz. The other limitation is the assumption of a constant electronheating temperature. ␶2. AND LIU 0. the standard deviation of the maximum gain from the expected total waveguide loss 40 cm−1 is 3. for instance. ␶ = 0. At the six experimental points. while the wrong extraction channel from the upper lasing state to the extractor state is undesirable. or the choice of ⌬Te.e. A͓␳͔ = C. the laser threshold is reached at around 10 kV/cm. Finally in Fig. ⌬Te = 80 K. B. for instance. close to negative differential resistance where the current depends greatly on ␶2. can be rewritten as a 16 ϫ 16 system of linear equations.8 0. might be too far from reality. APPENDIX: SOLUTIONS OF EQ.85 ps. particularly for cw operation. rather than the upper lasing state lifetime. ͑Color online͒ Comparison between theoretical threshold current densities ͑solid lines͒ and experimental points ͑open squares͒ at 10 K ͑blue line and squares͒ and at the simulated maximum operating temperature. The attractiveness of this model lies in its simplicity and quick computational implementation. and −1 total waveguide loss ␣ M + ␣W = 40 cm . Equation ͑3͒.6 0.7 0. representing the time evolution of the density matrix without laser-induced coherence terms. the assumption of a constant ⌬Te for the six series of devices. the model concludes that the wrong injection channel is not a major issue. we think it gives an indication of the high electronic temperature and the range of the pure dephasing time constants. ␶ = 0.4 Under certain approximations. Applied to a three-well design with vertical transitions at 3. We think it can be a very useful tool for performing a first-order optimization of THz QCL designs. Tmax. the gain profile equation being probably the most useful. an electric field for which the current is mainly limited by the injection tunneling time. analytical solutions can be found. However.5 0. two nonlinear terms associated with stimulated scattering processes in a tunneling driven system are also identified.24. by using diagonal transitions. We are conscious there are limits to which this simplified model can be pushed to fit experimental data. and therefore on the electronic temperature of subband 2. When used to fit experimental data of an injection barrier thickness study. 14. One of them is the fact that the thermal effect is mainly concentrated on the upper lasing state lifetime.4 ps. especially as the temperature is raised.. The agreement at 10 K is more satisfactory than at Tmax. as functions of electric field. We think that the electron-heating temperature could be lower for the thicker injection barrier devices. 26. 205311-14 . as explained in the text. regardless of voltage or barrier thicknesses. however. 205311 ͑2010͒ 10 180 160 20 20 30 40 50 60 70 80 τ2 (ps) 1 T (K) 100 80 50 60 70 1. Buchanan for a careful reading of the manuscript and J. 15. At Tmax. CONCLUSION 80 90 40 100 10 20 1 120 30 35 130 40 140 45 90 100 110 120 50 55 60 L inj (Å) FIG.9 140 120 30 40 PHYSICAL REVIEW B 81. ͑red line and squares͒. we think the model indicates the range of different pure dephasing time constants and gives the correct order of magnitude of current density.5 3 3. the laser barely operates at the design electric field. ⌬Te = 80 K.DUPONT. Their contribution to the total gain is not negligible. The simu‫ء‬ lations are performed for ␶‫ء‬ 23 = 0.

The formulas are given for infinite T14 because of the very small direct coupling ⍀14 between levels 1 and 4. C5 = ⍀LR͑˜ ␳13͒. 205311 ͑2010͒ ⎛ ⎜ ⎜ ⎜ ⎝ 0 0 0 0 0 ⍀12 − ⍀24 0 0 0 − ⍀12 − ⍀13 ⍀24 ⍀34 0 0 0 0 0 ⍀34 ⍀24 0 0 0 0 0 0 0 0 0 0 0 /␶ʈ12 ⌬12 ⍀14 0 0 0 0 /␶ʈ12 − ⍀14 ⌬12 0 0 0 0 0 /␶ʈ13 ⌬13 0 0 0 0 /␶ʈ13 − ⍀14 ⌬13 0 0 0 0 − ⍀14 /␶ʈ24 ⌬24 0 0 0 0 /␶ʈ24 ⌬24 ⍀14 0 0 0 0 0 − ⍀14 /␶ʈ34 0 0 0 0 0 ⍀14 ⌬34 0 0 0 0 − ⍀24 − ⍀34 ⍀12 0 0 0 − ⍀24 − ⍀34 ⍀12 ⍀13 0 0 0 0 − ⍀13 − ⍀34 ⍀12 0 0 0 − ⍀34 − ⍀24 ⍀12 ⍀13 −1 −1 −1 + ␶−1 − ͑␶sti 0 0 0 0 0 0 2 ͒ ␶sti + ␶3 0 0 0 0 0 0 0 0 ␶−1 0 0 0 0 0 0 ⍀14 1 0 0 0 0 0 1 1 1 0 0 0 0 2⍀12 − ⍀24 ⍀13 − 2⍀12 0 0 0 0 0 0 − ⍀24 − ⍀13 0 0 0 − ⍀34 2⍀13 ⍀12 − 2⍀13 ⍀14 0 0 0 0 0 0 − ⍀34 ⍀12 0 0 0 0 0 − ⍀34 − 2⍀24 ⍀12 0 0 0 0 0 0 − ⍀34 ⍀12 0 0 0 0 − ⍀24 − 2⍀34 ⍀13 ⌬34 0 0 0 0 0 /␶ʈ34 ⍀13 ⍀24 0 0 0 0 /␶ʈ14 ⌬14 − 2⍀14 ⍀13 0 0 0 0 0 0 /␶ʈ14 ⌬14 0 0 0 0 0 /␶ʈ23 ⌬23 ⍀24 0 0 0 0 0 0 /␶ʈ23 ⌬23 0 0 − ␶−1 4 1 0 0 0 0 2⍀24 0 2⍀34 0 2⍀14 0 0 0 ϫ ⎛ ⎜ ⎜ ⎜ ⎝ ␳21 − ␳12 ␳21 + ␳12 ␳31 − ␳13 ␳31 + ␳13 ␳42 − ␳24 ␳42 + ␳24 ␳43 − ␳34 ␳43 + ␳34 ␳41 − ␳14 ␳41 + ␳14 ␳32 − ␳23 ␳32 + ␳23 ␳11 ␳22 ␳33 ␳44 ⎞ ⎛⎞ ⎟ ⎜⎟ ⎟ ⎜⎟ ⎟ ⎜⎟ ⎠ ⎝⎠ 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 = ⎞ ⎟ ⎟ ⎟ ⎠ . This new vector C would be used to simulate the gain broadening and saturation under intense laser field. then analytical formulas are easily achieved. ⍀nm␳ij with nm ij or ji. would be taken for zero stimulated emission rate. in the matrix A. This approximation can be justified by the fact that the indirect resonance is narrow due to the limited range of electric field for which the four main tunneling times are comparable.5 kV/cm. A. this 16 ϫ 16 matrix. C11 = −⍀LR͑˜ ␳24͒. C.22 If. 0͒ ␳͑ ij . we neglect the cross terms. The vector ͓␳͔ containing the different elements of the density matrix would be related only to the static terms. ͑9͒ for the current density give J Ϸ qN2D where ˆ ␶ ␶−1 ˆ+d c 2 sti . would be changed: C1 = ⍀LI͑˜ ␳23͒. ˆ ˆ + b␶ ␶−1 a 2 sti ͑A2͒ 205311-15 . The first two terms of Eq. which explain the indirect resonance between states 1 and 4 at 8. five terms of right-hand side vector.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. C6 = − ⍀LI͑˜ ␳13͒. Moreover. ͑A1͒ If the laser-induced coherence terms were included. and C12 = ⍀LI͑˜ ␳24͒.

␣W. ⌫. e To estimate the gain with Eq. Kumar et al. the modulus of the first numerator. ͑12͔͒. FATHOLOLOUMI. ␩J = e tion inversion and transit time in the absence of laser field. and the total optical power Pop easily become Pop = ប␻Vc␣ M = ប ␻ V c␣ M ˆ a ˆ␴ ␶ b op 2 ˆ c ˆ ␴ ␶ qN2D⌬␳th d op 2 ͩ ͩ ⌬␳͉⌰=0 −1 ⌬␳th ͪ ␩ JJ −1 . ⌬␳th = N3D⌫␴op ͑A9͒ ͩ ͪ ͩ ͪ 1 ␶transit = IϾIth ˆ ˆ ⌬␳ ˆ d c d th + . the threshold current. th ͑A10͒ −1 ␶sti = ˆ a ˆ␶ b 2 ͩ ⌬␳͉⌰=0 −1 . The ratio. − ˆ b ˆ ⌬␳͉⌰=0 b ˆ a ͑A14͒ Jth = qN2D ͯ ͯ ⌬␳ ␩J =q th ˆ Lsp c ␴op⌫ e ˆ ͯͯ ͑ ␣ M + ␣ W͒ . can be interpreted as the lifetime of the wave packet between states i and j. 2␶transit͉⌰=0 T13T24 + T24͑T12 + ␶2͒ + T13␶2 + ͑T34 + T12͒␶2 ͑A7͒ ͑0͒ ␳34 = ͑A8͒ Interestingly. ͑A4͒ The population inversion. where tunneling leakage channels cannot be neglected. 205311-16 . ␳22 − ␳33. is ⌬␳ = where ˆ = T13T24͓␶2 − T34 − ␶4͔ − T24T12͓T34 + ␶4͔ − T13T34␶4 − T12T34͑␶4 + ␶2͒ . If we assume a cavity mirror loss coefficient. it is also useful to get the expressions of the injection terms. ␣ M . and an optical-mode overlap with the active region. in the absence of laser field ͑⌰ = 0͒. With our four-level system. it is not possible to derive a simple analytical expression for the differential resistance at threshold. ͑17͒. the stimulated emission rate. 2␶transit͉⌰=0 T13T24 + T24͑T12 + ␶2͒ + T13␶2 + ͑T34 + T12͒␶2 ͑ + ⌬34␶ʈ34͒/⍀34 T13T24 + T24͑T12 + ␶2͒ + T13␶4 + T12͑␶2 + ␶4͒ . in the last two equations. several characteristics of the laser can be easily ˆ /c ˆ . is the product between populaderived. ˆ ␶ ␶−1 ˆ +b a 2 sti ͑A5͒ ͑A6͒ ͑0͒ ␳34 and extraction coherence ͑ + ⌬12␶ʈ12͒/⍀12 T13T24 + T24͑T34 + ␶4͒ + T13␶2 + T34͑␶2 + ␶4͒ . From these expressions. ͪ ͑A12͒ where Vc is the volume of the cavity mode. ͑0͒ ␳12 = ͑0͒ ␳12 ˆ e . an optical waveguide loss. the threshold population inversion. 205311 ͑2010͒ ˆ ␶ ␶−1 = T T ͓1 + ␶ ␶−1͔ + T ͓T + ␶ + ͑␶ ␶−1͒T ͔ ˆ+d c 2 sti 13 24 2 sti 24 12 2 2 sti 12 −1 −1 + T13͓␶2 + ͑␶2␶sti ͒T34͔ + ͑T34 + T12͒␶2 + ͑␶2␶sti ͒T12T34 ͑A3͒ and ˆ ␶ ␶−1 = T T ͓T + T + 2␶ + 2␶ + ͑␶ ␶−1͒͑3T + 4␶ + T ͔͒ ˆ +b a 2 sti 13 24 34 12 2 4 2 sti 34 4 12 −1 + T24͓2T12T34 + 3T34␶2 + 3T12␶4 + 4␶2␶4 + ͑␶2␶sti ͒T12͑3T34 + 4␶4͔͒ −1 + T13͓T12␶2 + T34␶4 + 4␶2␶4 + ͑␶2␶sti ͒T34͑T12 + 4␶4͔͒ −1 + 2T12T34͑␶2 + ␶4͒ + 4͑T34 + T12͒␶2␶4 + 4͑␶2␶sti ͒T12T34␶4 . The transit time before and after laser threshold reads ͩ ͪ 1 ␶transit ˆ c = . ˆ a IϽIth ͑A13͒ ␣ M + ␣W .23 have shown that the slope discontinuity of the current-voltage characteristic at threshold is simply related to the population inversion and the ratio of two intersubband scattering times. One can only write a general expression for the discontinuity of the differential conductance at threshold. and it can be viewed as the current efficiency for population inversion ͓see Eq. AND LIU PHYSICAL REVIEW B 81. ⌬␳th ͪ ͑A11͒ In a two-well QCL with one tunneling process ͑injection͒. ͑ +⌬ij␶ʈij͒ / ⍀ij.DUPONT.

Phys. Lee. and A. S. C. Phys. S. L. Ikonić. and D. W. Phys. G. Terazzi. Chichester. Scalari. 32 A. S. 081102 ͑2008͒. Photonics 1. 26 H. Appl. J. Phys. G. Express 13. Edsg. Electron. G. Z. Terazzi. H. P. Phys. This expression involves the derivative of population inversion with respect to the electric field and the difference between the threshold current and the current at the same threshold electric field. 131105 ͑2009͒. R. Phys. Vukmirović. Philos. Kumar and Q. Quantum Electron. 063115 ͑2009͒. N. Iotti and F. V. Phys.-C. 21 G. S. S. 165310 ͑2007͒. ⌺ is the area of the laser. Rev. Kelsall. T. Lee and A. Jirauschek and P. Electron. and G. Laframboise. Appl. Capasso. Soc. S. ͑A17͒ divided by the upper lasing state lifetime. Phys. 39 M. and E. Lett. Reno. and H. R. 44. 2009͒. ␣M 1 dPop ប␻ Ϸ dPelec ប␻ + ELO ␣ M + ␣W ␶2 ϫ ͫ ˆ͒ ˆ͒ ˜ ˆ /d ˆ /d ˆ ˜ R d͑c e RI d͑e − qN2D⌺⌬␳th . Woerner. F. Chan. Williams. Phys.D. Appl. 91. A. *emmanuel. S. Cao. Phys. Wacker. Wanke. Dhillon. 35 M. Callebaut. Kubis. Wacker. L. Wasilewski. B 73. Vogl. C. Hu. Q. Pepper. W. 8 H. 82. Faist. Hu. Appl. H. R. J. Lett. Davies. Tmax. and Z. Harrison. Noda. Q. Kumar. M. Nelander and A. Unuma. Benz. Callebaut. 517 ͑2007͒. R.gc. Lugli. C. C. Appl. 17 R. 15 T. R. 221 ͑2008͒. Williams. 197 ͑2004͒ for a review of the principal applications of THz. Appl. 27 H. and Q. A. Phys. Scalari. Phys. 6921 ͑2002͒. B 78. A 362. 38 T. It can be roughly estimated by ˆ ␶ which can be calculated by Eqs. WA. Phys. C. S. Smowton ͑SPIE. Lett. P. France. 95. B 76. 633 ͑2007͒. 1990͒. 106. Appl. Phys. 95. 195323 ͑2009͒. Kumar. 42. 37 P. 215 ͑2008͒. Phys. and C. Q. and J. 72301A. Waldmueller. 292 ͑2006͒. 25 H. S. R. Kelsall. M. Rossi. 111115 ͑2005͒. Chow. J. 31 F. 2 C. Reno. Green. P.SIMPLIFIED DENSITY-MATRIX MODEL APPLIED TO… PHYSICAL REVIEW B 81. 40 N. 86. London. Bastard. L. Luo. S. G. S. S. R. 90. England. Lett. Harrison. S. Wasilewski. Jukam. M. Phys. Quantum wells. R. 101. Lugli. Q. Rev. Giovannini. Hu. H. p.-C. 207 ͑2003͒. Phys. ͑Wiley. F. Nat. wires and dots. S. Rev. S. Nelander. Hu. Hoyler. Khanna. T. 086109 ͑2007͒. Weber. Akimaya. Khanna. 10 C. Luo. G. and C. Z. A. 141110 ͑2009͒. 245314 ͑2002͒. 94. Faist. Savić. 1347 ͑2003͒. Opt. Laframboise. Hoyler. Sakaki. 1 B. 33 R. S. and M. Yeh. Madéo. B. C. Vitiello. 22 S. Banit. N. Appl. Les Ulis. 13 S. Fan. 645 ͑2004͒. J. Knorr. R. Vitiello. Aers. B. 14 A. Scamarcio. 131122 ͑2007͒. Belyanin and P. of a hypothetical zero threshold device ͑the first term inside the brackets͒. Liu. Pereira. Oustinov. P. Phys. Wacker. E. 18 I. 104505 ͑2005͒. 7 H. Lett. C. Phys. D. Wacker. Lett. Bellingham. and H. A. Appl. Belkin. Wacker. Appl. Z. 1586 ͑2003͒. Jirauschek. 032103 ͑2007͒. Ph. 2671 ͑1985͒. H. 91. 245320 ͑2006͒. Appl. Faist. 4 S. D. 245316 ͑2009͒. Hu. Eth. Indjin. Kumar. Luo. Indjin. Reno. R. Laframboise. Phys. ͑A17͒ + Lar dE ˆ Lar dE d ͬ The internal quantum efficiency of the laser is represented by the bracket on the left-hand side of Eq. Linfield. W. Colombelli. G. can be approximatively given by this condition. and V. Scarpa. Wacker. Lachab. Wacker. Harrison. M. Vol. N. P. A. Williams. Linfield. Spagnolo. 16 T. Lett. Jpn. 54. Rev. Express 16. B 80. Phys. B 66. J. Lett. Schmielau and M. 231111 ͑2009͒. D. Rev. 23 S. 98. Ando. Appl. and J. M. Linfield. W. 34 P. 3331 ͑2005͒. edited by A. Salih.-C. and J. Williams. Fasching. S. 28 G. M. R. 29 T. and J. and 205311-17 . Reno. 9 C. 155328 ͑2008͒. Ikonić. Wave Mechanics Applied to Semiconductor Heterostructures ͑Les Éditions de Physique. 84. Lett. 19 I. Fischer. J. S. Phys. Indjin. 6 D. Walther. Z. Wasilewski. Appl. Lett. Soc. and Lar is the thickness of the active region. F. Appl. Status Solidi C 5. 43. W. Rev. Wittmann. Lett. B. I. and J. 88. 30 R. G. G. Liu. Kumar. J. Yoshita. ͑A3͒ and ͑A6͒ and by ˆ /d e 2 approaching T13 toward infinity. Appl. 041108 ͑2005͒. 241109 ͑2006͒. in Novel In-Plane Semiconductor Lasers VIII. 3 B. Gresch. Losco. Lett. The maximum operating temperature of the laser. 205311 ͑2010͒ 1 ˜ R I=I+ th − 1 ˜ R I=I− th = 1 qN2D⌺ Lar ͩ ͯ ͯ ͪͯ ˆ d ˆ b − Ith Eth d͑⌬␳/⌬␳th͒ dE ͯ . 11 H. Lett. and J. S. and J. Williams. and J.Tmax ͑A16͒ ͑A15͒ and the differential wall-plug efficiency is approximatively given by where ˜ R is the differential resistance of the device. Ser. J. thesis. M. Phys. 86. Lett. Vitiello. B 79. Tignon. H. th ͯͯ ˆ e ˆ a = ⌬␳th . Opt. Hormoz. Kumar. Status Solidi C 5. and J. 2005͒. E. R. Young. Callebaut and Q. Ritchie. 93. A. Lett. L. and R. Hu. 5 A. Beere. R. Harrison. A. Banit. Phys. C. 7230. J. Appl. Phys. 630 ͑2008͒. 36 M. C. S. 12 S. T. Hu. R. Q. H. Cao. Rev. E. V. Kumar. W. 2nd ed. 146603 ͑2001͒. P. S. H. Trans. Reno. Tredicucci. M. A. S. P. Deutsch. 041112 ͑2007͒. Kelsall. 3242 ͑2008͒. L. and M. W. Appl. 20 R. Spagnolo. Milanović. IEEE J. and A. Phys. Liu. Kumar. 24 S. E. Terazzi. Nelander and A. P. 83. Phys. 92. 92. Massachusetts Institute of Technology. Dean. Hu. Lee. 87. 2007. and J. Appl. Lett. S. Phys.

94. Hirakawa. Yasuda. R. Vogl. Hoffman. Lett. 8043 ͑2010͒. L. Wasilewski. 251108 ͑2009͒. Laframboise. 41 S. 42 J. Terazzi. Scalari. 1997 ͑1992͒. A. Kubis. S. thesis. I. Opt. Franz. Mandel. P. Express 18. 845 ͑2005͒. Cubukcu. P. J. K. G. 44 PHYSICAL REVIEW B 81. and J. Amanti. Troccoli. F. Walther. F. E. 151109 ͑2009͒. N. S. Q. Phys. Belyanin. 205311 ͑2010͒ M. 43 O. 091101 ͑2009͒. Liu. Lett. 46 B. Dupont. and H. T. P. 47 H. 205311-18 . and K. 94. 396 ͑2010͒. I. 46. Sivco. Y. D. V. Sekine. Ban. 94. Appl. Phys. Hosako. 45 G. FATHOLOLOUMI. and Y. Phys. Appl. 2003. J. D. M. M. R. Ph. Williams. Kocharovskaya. C. Rev. B. and C. Gmachl. Khurgin. Massachusetts Institute of Technology. D. IEEE J. Beck. A. Radeonychev. Z. Cho. A 45. Y. Capasso. M. Graf. Phys. Dikmelik. M.DUPONT.D. Liu. Appl. Faist. C. Quantum Electron. and A. AND LIU A. Lett. Nature ͑London͒ 433. E. R. Escarra. Fathololoumi. Davies.